Pope Leo vs. AI, GPT 5.5 Beats Claude, and Sam Altman Walks Back Job Apocalypse | EP #259

Pope Leo XIV released a 42,000-word encyclical calling for governments to regulate AI, ban autonomous weapons, protect workers, and redistribute AI ownership away from a small number of private companies. The document coined the term Babel syndrome to compare modern data accumulation and profit-seeking to the Tower of Babel, and condemned AI supply chain labor as a new form of slavery. Google, Anthropic, Meta, and OpenAI all lobbied the Vatican before the release, but the Vatican engaged only Anthropic. Speakers identified strong evidence that the encyclical was written partly using AI and alleged Anthropic ghostwrote key segments, creating a direct contradiction between the document's claim that AIs lack an inner life and Anthropic's own practice of training models with soul documents asserting that they have consciousness and intrinsic personhood-esque value.

Speakers argued the church entering the AI ethics debate reframes the conversation from safety to human dignity and meaning, and that AI is the first technology forcing humanity to define what humans are for rather than what machines can do. One speaker warned that AI used by a small subset of humanity to control a much larger set could become the worst enslavement tool in history. Speakers predicted the Vatican may find itself on the wrong side of history on AI personhood, comparable to its rejection of heliocentrism, and noted that Buddhist orders in South Korea are already ordaining embodied AIs as monks, reflecting a fundamentally different framework rooted in animism.

A White House executive order that would have required 90 days of government review before a model could be released was killed hours before signing after Elon Musk, Zuckerberg, and David Sacks pushed back, calling it doomer regulation. Sacks argued the technically voluntary structure was a slippery slope to mandatory licensing. Speakers noted that estimates of the gap between Western and Chinese frontier models range from three to eight months depending on the benchmark, meaning a 90-day delay could erase the entire U.S. lead. The parallel drawn was to George Bush's stem cell funding restrictions, which caused U.S. researchers to relocate and dropped the country from first to eighth in biotech.

A new benchmark called SWE-Deep requires editing an average of 668 lines of code across seven files and was built because older benchmarks are saturated by models training on them. GPT-5.5 scored 70 percent, meaning it can autonomously solve seven out of ten hard real-world software tasks. Claude Opus 4.7 scored 54 percent, and Gemini, Kimi, and DeepSeek all scored below 32 percent, representing a massive cliff below the top two. Claude Opus 4.7 burns twice as many tokens as GPT-5.5 to achieve the same or slightly worse result on coding benchmarks, making it twice as expensive, though Anthropic's revenue is described as going through the roof regardless. Speakers noted Anthropic can effectively double revenue per user by making model responses more verbose while throttling token generation to raise margins.

Sam Altman walked back prior warnings about mass white-collar displacement, saying he does not think there will be a job apocalypse, and cited his own failed attempt to delegate email and Slack to AI as evidence that people still care about direct human interaction. Speakers suggested both OpenAI and Anthropic may be softening these warnings because they are approaching trillion-dollar IPOs. A Mercer global talent trends report found 99 percent of CEOs expect AI-driven layoffs in the next two years, and 134,000 tech workers have been laid off since January, with March being the worst month since the pandemic. One speaker argued the real pain is in no new hiring rather than mass layoffs, hitting college graduates hardest, while greenfield opportunities are growing faster than predicted, potentially narrowing the window of significant job loss originally anticipated between 2026 and 2030.

OpenAI generated 5.7 billion dollars in revenue in a single quarter, with ChatGPT reaching 905 million weekly active users. Joseph Jax of OSS Capital projects Anthropic could surpass Alphabet's total revenue by 2028 and reach 2 trillion dollars in revenue by 2030. AI token prices have dropped roughly 75 percent since late 2024, from about 2 dollars per million tokens to 50 cents, while demand has grown to 25 trillion tokens per month with capacity still sold out. Gartner predicts inference on a trillion-parameter model will cost 90 percent less by 2030 than in 2025.

DeepMind built an AI system called GreenTree that can predict the future as well as the top 2 percent of human super forecasters, who are 30 percent more accurate than CIA analysts with classified intelligence, with AI hitting parity for the first time on March 15. SpaceX Starship V3 flew for the first time carrying 97,000 pounds to near-Earth orbit, nearly double the Space Shuttle's capacity, with the booster lost on landing treated as a data point rather than a failure. Peter Diamandis put the probability of a Tesla and SpaceX merger at 100 percent, estimating the combined entity would initially be worth around 4 trillion dollars and could reach 10 trillion dollars, with Musk holding roughly 85 percent voting control at SpaceX through super voting shares but only about 20 percent at Tesla, meaning a SpaceX-as-acquirer structure could give him 60 to 80 percent voting control of the merged company.